Fast Cross-Validation for Incremental Learning

Authors: Pooria Joulani, Andras Gyorgy, Csaba Szepesvari

IJCAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments with stateof-the-art incremental learning algorithms confirm the practicality of the proposed method.
Researcher Affiliation Academia Pooria Joulani Andr as Gy orgy Csaba Szepesv ari Department of Computing Science, University of Alberta Edmonton, AB, Canada {pooria,gyorgy,szepesva}@ualberta.ca
Pseudocode Yes Algorithm 1 TREECV s, e, ˆfs..e
Open Source Code No The paper does not provide explicit statements or links to open-source code for the described methodology.
Open Datasets Yes We used datasets from the UCI repository [Lichman, 2013], downloaded from the Lib SVM website [Chang and Lin, 2011].
Dataset Splits Yes The general recipe for computing CV estimate is to run a learning algorithm separately for each CV fold, a computationally expensive process. ...k-fold cross-validation (k-CV): the dataset is partitioned into k subsets of approximately equal size, and each subset is used to evaluate a model trained on the k 1 other subsets to produce a numerical score; the k-CV performance estimate is then obtained as the average of the obtained scores.
Hardware Specification Yes The tests were run on a single core of a computer with an Intel Xeon E5430 processor and 20 GB of RAM.
Software Dependencies No The algorithms were implemented in Python/Cython and Numpy. No specific version numbers for these software dependencies are provided.
Experiment Setup Yes The regularization parameter was set to λ = 10 6 following the suggestion of Shalev-Shwartz et al. [2011]. For LSQSGD, we used the UCI Year Prediction MSD dataset (463,715 data points, 90 features) and, following the suggestion of Nemirovski et al. [2009], set the step-size to α = n 1/2.